Publication number | US6928434 B1 |
Publication type | Grant |
Application number | US 09/867,801 |
Publication date | Aug 9, 2005 |
Filing date | May 31, 2001 |
Priority date | Jan 31, 2001 |
Fee status | Paid |
Publication number | 09867801, 867801, US 6928434 B1, US 6928434B1, US-B1-6928434, US6928434 B1, US6928434B1 |
Inventors | Lawrence J. Choi, Christopher B. Kuenne, E. Holstein II Kurt |
Original Assignee | Rosetta Marketing Strategies Group |
Export Citation | BiBTeX, EndNote, RefMan |
Patent Citations (7), Non-Patent Citations (2), Referenced by (63), Classifications (15), Legal Events (16) | |
External Links: USPTO, USPTO Assignment, Espacenet | |
This application relates to, claims priority to, and incorporates by reference herein in its entirety, the following pending U.S. patent application:
This invention relates to and incorporates by reference herein in their entirety, the following pending U.S. patent applications:
The present invention relates at least to the field of statistics, and, more particularly, to a method and system for clustering optimization and applications.
The invention will be more readily understood through the following detailed description, with reference to the accompanying drawings, in which:
Introduction
The combination of Wall Street pressure to accelerate earnings growth and an ever-increasing fragmentation of consumer preferences and media habits has created a strategic imperative for businesses to identify and understand a panoply of consumer product and service choices through marketing tools and frameworks, most notably customer segmentation.
Traditional approaches to segmentation have at least the following three fundamental flaws:
Embodiments of methods and systems of the present invention can solve each of these three flaws through a unique and distinctive set of business processes and econometric modeling techniques and thereby provide businesses with the ability to create breakthrough marketing initiatives. Moreover, these inventive processes and techniques can be extended to solve numerous problems outside of the fields of marketing and business.
Why the Rosetta Inventive Method Works
Rosetta's inventive methodology combines two seemingly disparate disciplines, microeconomics and psychology, under the backdrop of econometrics.
Microeconomics teaches that purchase and sales behaviors are logical, rational, and predictable, but provides no insight into the customer beliefs that drive brand choice.
Psychology teaches that a finite set of personality types exist in a given population but provides no insight into the decision rules that lead to brand choice.
Econometrics provides the quantitative and analytic rigor needed to identify the finite sets of personalities each with unique and predictable purchase/sales behavior patterns and decision rules.
The Rosetta Segment-Based Marketing System combines and provides the benefits of the insights of economics and psychology in a quantitatively rigorous framework with rapidly actionable applications.
The Rosetta Difference
Rosetta's unique segmentation process, beginning with data collection and culminating in clustering optimization, advantageously defines segments that are collectively more differentiated and individually more homogeneous than those developed using previous industry best practices.
An example of these improved segments is shown in
The solid line of
An important criterion for evaluating a segmentation solution is determining whether response patterns for generated clusters are statistically different from the response pattern of the overall population of respondents. As call-out box 1 of
Another important criterion for evaluating segmentation structures is logical consistency. In call-out box 2, the k-means segment is revealed to have a logical discrepancy. For variables that probe a respondent's reliance on the medical establishment (i.e. physicians, pharmacists, nurses), the k-means segment indicates that the segment agrees with being reliant on the medical establishment for treating illnesses. Logically, the same k-means segment should disagree with variables that indicate self-reliance for treating illnesses. However, the same k-means segment strongly agrees with the self-reliance for treating illnesses variables. The Rosetta segment is, in contrast, logically consistent because it agrees with being reliant upon the medical establishment for treating illnesses and disagrees (i.e. very few segment members agree) with being self-reliant for treating illnesses.
An additional advantage of the Rosetta segmentation approach is that more variables are revealed as informative regarding each segment's beliefs toward a particular concept. If a segment's response distribution for a set of variables closely mirrors that of the total population, the variables in question are generally dismissed as either non-discriminating (because there is no statistical variance from the overall population) or as price of entry (i.e. a concept to which so many individuals in the population respond identically that the concept is considered an existential to the market in question).
Rosetta's approach, however, reveals variables to be discriminating where other approaches, such as k-means clustering, fail. Call-out box 3 reveals how response to variables related to patient compliance with physician instructions (e.g. “get more bed rest”, “drink more fluids”, “take medication for 10 days”) are not statistically different from the total population for the k-means segment. These variables would normally be ignored as non-discriminating or dismissed as price-of-entry. Rosetta's approach reveals that these compliance variables are indeed highly discriminating in defining the segment. In addition, it makes sense logically for Rosetta's segment to agree with being reliant on the medical establishment for treatment AND to agree with variables relating to compliance. However, price-of-entry variables do exist in virtually all market categories (e.g. in carbonated sodas, “my soda needs to be carbonated”, is price-of-entry). A segmentation structure that discovers NO price-of-entry variables is suspect. Call-out box 4 is an example of where the Rosetta approach has indeed discovered some price-of entry variables related to product features.
Overview
The Rosetta Segment-based Marketing Process includes several inventive techniques that are disclosed herein. They include:
A description of marketing-oriented applications for each of these techniques follows, the description including background information, a brief problem statement, a brief solution statement, a discussion of the impact and/or other applications for the technique, and finally, a brief description of the methodology of the technique. Following these descriptions of these techniques as applied to marketing, a description of flowcharts showing exemplary embodiments of the techniques as applied more generally is provided.
Throughout these descriptions, several terms are used synonymously. For example, the term “segment” is synonymous with the term “cluster”, which is a grouping of similar observations. Likewise, the term “clustering structure” is a synonym for “clustering solution”, either of which are a finite set of clusters for a dataset, with, in some cases, constituents that are mutually exclusive and collectively exhaustive (“MECE”).
Also, the term “respondent” is synonymous with the term “observation”, which can be viewed as a record (or row) in a dataset for which values are recorded for a particular variable. Further, the term “question” is synonymous with the term “variable”, either of which can be viewed as a field (or column) in a dataset, the field having a recorded value for each observation. Moreover, the term “possible answer” is synonymous with the term “possible value”, and the term “provided answer” is synonymous with the term “provided value”.
I. Babbitt Score
A. Description
B. Impact/output
Babbitt Score | Evaluation | ||
≦50 | Keep | ||
50 < x ≦ 100 | Further Analysis | ||
>100 | Eliminate | ||
C. Methodology/components
Top 2 Box
Response | 5-Point Scale | 7-Point Scale | Description |
Agree/Important | 1, 2 | 1, 2 | Top 2 Box |
Neutral/No Opinion | 3 | 3, 4, 5 | Middle 1 or 3 |
Disagree/Unimportant | 4, 5 | 6, 7 | Bottom 2 Box |
Top 3 Box
Response | 7-Point Scale | Description |
Agree/Important | 1, 2, 3 | Top 3 Box |
Neutral/No Opinion | 4 | Middle |
Disagree/Unimportant | 5, 6, 7 | Bottom 3 Box |
A. Description
B. Impact/output
C. Examples:
C. Methodology/components
A. Description
B. Impact/output
Cycle N | Cycle N + 1 | |||
Conventional | 1,000 iterations | 3,000 iterations required | ||
Winning solution | 3,000 iterations executed | |||
The inventors | 1,000 iterations | 3,000 iterations required | ||
Winning solution | 2,000 iterations executed | |||
C. Methodology/components
A. Description
B. Impact/output
C. Methodology/components
A. Description
B. Impact/output
C. Methodology/components
A. Description
B. Impact/output
C. Methodology/components
A. Description
B. Impact/output
C. Methodology/components
A. Description
B. Impact/output
C. Methodology/components
D. Data Types
Some of the disclosed processes use scalar, categorical, and/or continuous variables from surveys fielded to a study population and/or behavioral/purchase data obtained from either the business itself or a panel, such as IRI, AC Nielsen, comScore, Simmons, MRI, or Nielsen ratings panel (not affiliated with AC Nielsen).
Data Input | Data Type | Accessibility |
Needs | Scalar or Categorical | Common; used by most |
Variables | industry firms | |
Attitude | Scalar or Categorical | Common; used by most |
Variables | industry firms | |
Behavior | Scalar or Categorical or | Common; used by most |
Variables | Continuous transformed into | industry firms |
Categorical/Scalar | ||
Panel- | Categorical or Continuous | Uncommon; requires |
Derived | transformed into | subscription or alliance with a |
Variables | Categorical/Scalar | panel company with IRI or |
Nielsen. The inventors have | ||
relationships with IRI and | ||
comSCORE | ||
Business- | Categorical or Continuous | Common; used by most |
Derived | transformed into | industry firms |
Variables | Categorical/Scalar | |
Composite | Categorical or Continuous | Rare; created by the inventors |
Variables | transformed into | using database-derived and/or |
Categorical/Scalar | panel-derived variables | |
At activity 1100, for all observations (respondents), provided values (e.g., responses) associated with a variable (e.g., survey question), can be converted into proxy values, if necessary to insure that each variable has only a finite set of values to be analyzed. For example, if the provided values are continuous, those provided values can be converted into one of several scalar/categorical or discrete proxy values. By way of further example, if a variable was “On a scale of 1 to 10 (with 10 being best), how do you rate the service you received?”, and a provided value was 8.2, that provided value could be converted to a proxy value of 8. Furthermore, proxy values can be a subset of provided values. For example, if a provided value was categorical (e.g., “red”, “green”, or “blue”), then the proxy values can also be categorical, and can be identical to the provided values. Moreover, the conversion of provided values to proxy values explained in relation to activity 1100 applies to the provided values of Methods 2 through 8 and 10 as well (described below).
At activity 1200, proxy values associated with a variable (e.g., survey question) can be segregated into categories, (e.g., Top 2 box, Top 3 box , etc.). At activity 1300, for each category, a response distribution can be determined by dividing a number of responses in the category by the total number of corresponding responses for all categories. At activity 1400, a top box score can be calculated by adding the top box response (%) to the bottom box response (%) and subtracting an ideal distribution of neutrals (%) to obtain a result. Then, an absolute value of the result can be multiplied by 100.
At activity 1500, a difference score can be calculated by subtracting the bottom box response (%) from the top box response (%) and multiplying an absolute value of the result by 100. At activity 1600, an effectiveness score, also known as the Babbitt score, for the survey questions can be calculated by adding the top box score to the difference score. At activity 1700, the survey question can be evaluated based on the effectiveness score.
The first technique can be shown at activity 2350, where a specified initial cluster assignment can be obtained from a previous dataset.
The second technique can be shown at activities 2360 through 2362, where a systematic search can be made for the initial cluster assignment. In this systematic search, a pair of variables can be identified that creates a clustering solution that maximizes fitness score using the specified number of clusters. At activity 2361, the one variable that creates a clustering solution that maximizes score using the specified number of cluster can be discovered. At activity 2362, the discovered variable from activity 2361 can be held constant and a second variable that creates a clustering solution that maximizes fitness score using the specified number of clusters can be discovered.
The third technique can be shown at activities 2370 through 2373, where a thorough search can be made. During the thorough search, any two variables that together create a clustering solution that maximizes fitness score using the specified number of clusters can be identified. At activity 2371, the one variable that creates a clustering solution that maximizes fitness score using specified number of clusters can be discovered. At activity 2372, holding the discovered variable from activity 2371 constant, a second variable that in tandem with the variable being held constants creates a clustering solution that maximizes fitness score using a specified number of clusters can be discovered. At activity 2373, the second discovered variable can be held constant and a third variable can be found that in tandem with the second variable being held constant creates a clustering solution that maximizes fitness score. Activities 2372 and 2373 can be repeated interactively and iteratively by cycling through all possible combinations of variable pairings until the fitness score that can be calculated in activity 2373 can be maximized.
After one of the three techniques are followed for developing an initial cluster assignment, at activity 2400, the observation identifications, the cluster identifications, and the fitness scores can be stored.
Referring to
At activity 2510, one random observation from a random cluster can be selected and its cluster assignment can be changed. At activity 2520, the fitness scores of activity 2510 can be calculated. At activity 2530, the fitness score of activity 2520 can be compared with the fitness score of activity 2505. At activity 2531, if the fitness score of activity 2520 can be less than or equal to the fitness score of activity 2505, all possible cluster assignments for the observation selected in activity 2510 can be cycled through until the fitness can be maximized.
At activity 2532, the maximum fitness score of activity 2531 can be compared with that of 2505. At activity 2533, if the maximum fitness score of activity 2531 can be less than or equal to the fitness score of activity 2505, then the selected observation can be returned to the original cluster assignment. At activity 2534, method 2 returns to activity 2510.
Considering again the comparison at activity 2532, if the maximum fitness score of activity 2531 can be greater than the fitness score of activity 2505, at activity 2535 method 2 proceeds to activity 2540. At activity 2540, the observation identification can be stored as well as the new cluster assignments. At activity 2550, the cluster assignments of activity 2505 are replaced with the cluster assignments of activity 2540 and method 2 returns to activity 2510.
Considering again the comparison at activity 2530, if the fitness score of activity 2520 can be greater than the fitness score of activity 2505, method 2 proceeds to activity 2540. If method 2 can be looping from activity 2550, once the fitness score of activity 2520 can be compared with that of activity 2505 at activity 2530, method 2 then proceeds to activity 2551. At activity 2551, if the fitness score of activity 2520 can be equal to that of activity 2505, method 2 proceeds to activity 2560. At activity 2552, if the fitness score of activity 2520 can be less than or greater than that of activity 2505, method 2 returns to activity 2540.
Referring now to
At activity 2580, the iteration identification from activity 2570 can be compared with the total number of iterations specified in activity 2200. At activity 2581, if the iteration identification from activity 2580 can be less than the total number of iterations specified in activity 2200, method 2 can return to activity 2470. At activity 2582, if the iteration identification from activity 2580 can be equal to the total number of iterations specified in activity 2200, method 2 can proceed to activity 2600. At activity 2600, the iteration identification that produced the maximum fitness score, the maximum fitness score itself, the observation identifications, and the corresponding cluster assignments can be stored. At activity 2610, the data from activity 2600 can be placed in a file, such as an ASCII, .csv, .txt, or .prn file. At activity 2620, method 2 can be terminated.
At activity 4010, for each observation, a dataset is obtained having a cluster assignment for the observation and having a proxy value for each of the variables in the dataset, each variable having possible values. At activity 4020, for each observation, an estimate is made that a purposeful probability (a measure of a probability that an observation in a particular cluster provides an answer to a question in a non-random manner) that a particular possible value for a particular variable will be provided by observations assigned to a particular cluster. At activity 4021, probability variables for each cluster, variable and answer combinations are created as P_{m}(k, l). At activity 4022, the probability that answer value l is given by the observations in cluster m for variable k that has L_{k }possible answers is estimated. At activity 4023, a value δ is defined within a constraint that allows for usable output. The value
is a value that produces meaningful results. If δ=0, the resulting over-precision of the calculated probabilities can compromise computational efficiency.
At activity 4024, a computational process is executed across all
where N_{m}=the total number of observations in cluster m; N_{m}(k, l)=the number of observations in cluster m who give the l-th answer value to variable k; and
At activity 4025, for each observation, the purposeful probability can be stored and/or outputted. At activity 4030, for each observation and each possible value, a serendipity probability (a measure of a probability that a observation in a particular cluster will be associated with any of the possible values for a particular variable) can be calculated. If an observation i in cluster m selected responses “randomly”, then the probabilities of selecting his responses should be described as
At activity 4035, for each observation, a ratio of purposeful probability to serendipity probability can be calculated. At activity 4040, for each observation, a logarithm of the ratio from activity 4040 can be calculated to obtain a composition analysis score. At activity 4045, for each observation, the composition analysis score can be stored and/or outputted.
At activity 4050, for each observation, an assumption can be made that before an observation is made, the observation has an equal probability of being in any cluster. At activity 4055, for each observation, an assumption can be made that the purposeful probabilities are true. Thus, if observation i in cluster m purposefully and logically selected his responses, then the probabilities of selecting his responses should be described by P_{m}(k, l). At activity 4060, for each observation, a Bayes probability can be calculated that a particular observation can be in each cluster is conditional upon the observation s proxy value. At activity 4065, for each observation, the Bayes probability can be stored and/or outputted.
At activity 4080, for each observation, a percent of proxy values for the variables that equals a modes of that observation s cluster s proxy values for the corresponding variables can be calculated. At activity 4085, for each observation, the calculated percent can be stored or outputted.
At activity 4090, each observation can be classified based on the results obtained in activity 4045, 4065 and/or 4085.
At activity 5100, a dataset for observations can be obtained, the dataset having variables, possible values, provided values, and corresponding cluster assignments. For the purposes of
At activity 5210, if no weights are needed, method 5 can proceed to activity 5300. At activity 5220, if weighting can be needed, each cluster can be assigned a weight using the decision rules of one of activities 5221, 5222, or 5223. In the weighted situation, it can be preferable to assign weights to all clusters regardless of the magnitude of the clusters. In activity 5221, if the cluster can be unweighted, the weight can be set as w=1. At activity 5222, if the cluster can be to be overweighted, the weight can be set as w>1. At activity 5223, if the cluster can be to be underweighted, the weight can be set as 0<w<1.
At activity 5300, the clustering solution developed using Method 2 can begin to be approximated by developing a clustering solution that employs only 1 variable from {K} in each. In other words if there are K variables, then K optimized solutions can be created 1 per k in {K}. An optimized solution is a clustering solution that has achieved a maximum score (however defined) within its defined constraints (e.g. number of variables, number of respondents, number of iterations, number of clusters, etc.) At activity 5310, a dummy variable can be created for each cluster so that if there were M clusters, then there are M variables. For example, let Y_{im }designate a dummy variable for observation i in cluster m. Let i be a member of the set of observations {R}, k be a member of the set of variables {K}, and m be a member of the set of clusters {M}. At activity 5320, m variables can be populated per observation per clustering solution. At activity 5321, if observation i can be in cluster m then Y_{im }can be set to 1. At activity 5322, if observation i can be not in cluster m then Y_{im }can be set to 0. At activity 5330, all values for M variables can be stored for all observation for each clustering solutions.
Referring to
At activity 5350, for each observation i and variable k, L_{k }variables can be populated per clustering solution. Thus the total number of variables created per observation per cluster is K(L_{k}). At activity 5351, if observation i gives the l-th answer for variable k, then χ_{ilk }can be set to 1. At activity 5352, if observation i does not give the l-th answer for variable k, then χ_{ilk }can be set to 0.
At activity 5360, all values for K(L_{k}) variables for all observations per clustering solutions can be stored. At activity 5370, for each cluster in {M}, ordinary least squares can be used to regress all Y_{im }for all observations in {R} per clustering solution. Thus regression occurs on {1, {χ_{ilk}: kεK,1≦l≦L_{k}−1}} so that a linear approximation can be generated to the probability that an observation with a particular set of answers to the variables in {K} can be in a particular cluster within {M}.
At activity 5380, a simpler clustering solution can be constructed to the one generated using Method 2. By “simpler” what is meant is an approximation of the actual clustering solution using the specified constraints in this case using only one variable k within {K}. At activity 5381, if weights were specified in activity 5200, an index for each observation's cluster association can be created. An index can be created by multiplying the linear approximation to the probability of an observation's cluster assignment by that cluster's specified weight. An index is created for each possible cluster assignment. At activity 5383, each observation {R} can be assigned to the m-th cluster in {M} that gives the maximized index value of the clusters in the clustering solution.
At activity 5382, if weights were not assigned in activity 5200, each observation in {R} can be assigned the m-th cluster in {M} that gives the maximum value of the linear approximation to the probabilities of being in any of the cluster members of {M} as calculated in activity 5370 for each clustering solution. At activity 5390, the outputs of regression, (i.e., the coefficients and constants) can be stored as well as the variable identifications, observation identifications, the approximated cluster assignments, and the actual cluster assignments for all clustering solutions.
Turning now to
At activity 5420, clustering solutions can be approximated using only 2 variables in each. At activity 5421, variable k from activity 5411 can be held constant and activities 5310 through 5400 can be executed for all possible pairs comprising k from activity 5411 and the (k+1)-th variable.
At activity 5422, the winning 2 variable solution from activity 5421 can be refined. At activity 5423, the (k+1)-th variable from activity 5421 can be held constant and activities 5310 through 5411 can be executed for all possible pairs comprising the (k+1)-th variable from activity 5421 and the remaining variables in {K}, excluding the (k+1)-th variable and the variable k identified in activity 5411. The pattern in activity 5423 can increase serially as the number of variables used to approximate the clustering solution from Method 2 increases serially.
At activity 5430, a continual loop through activities 5420 through 5423 can be performed, sequentially increasing the number of variables used in activity 5420 at the beginning of each loop until a maximum solution (in terms of accuracy) is identified for a simpler clustering solutions that uses k_{max(x) }variables to approximate the clustering solution identified in Method 2. Therefore if the (k+2)-th variable is added to the pair of the k-th and (k+1)-th variables to create a clustering solution that best approximates the objective function (i.e., the original clustering structure developed in Method 2), then in the refining activity 5422, the (k+2)-th variable is held constant while the k-th and the (k+1)-th variables are replaced with all remaining variables to test the triplet of variables that best approximates the results of Method 2.
At activity 5440, the outputs of regression, the variable identifications, the approximate cluster assignments (and the corresponding observation identifications), the actual cluster assignments (and the corresponding observation identifications), and accuracy scores for only the maximized solutions for all solutions created up through and including k_{max(x) }variables can be selected and stored. At activity 5450 the stored information from activity 5440 can be placed into a file of any appropriate format, e.g. ASCII, .txt, .csv, and/or .prn. At activity 5460 Method 5 can be terminated.
At activity 6100, for each observation, a dataset is obtained, each dataset having variables, possible values, provided values, and corresponding cluster assignments (which could have been developed using Method 2). The dataset can consist of any combination of scalar, categorical, and/or continuous variables. At activity 6200, all continuous variables can be transformed into categorical or scalar forms. This transformation can be done by analyzing distribution boundaries within a series of ranges to find the boundaries that create as normal distributions as possible. In many situations, linear optimization is the most efficient method for performing this boundary analysis.
At activity 6300, the dataset can be refined to facilitate further analysis. At activity 6310, if the dataset has 100 or fewer variables, Method 6 can proceed to activity 6400. At activity 6320, if the data set has greater than 100 variables, the dataset can be reduced as much as possible. To perform this reduction, at activity 6321, any of the following analytical techniques can be implemented: log scores, tree analysis, regression, and/or discriminant analysis. These analytical/statistical techniques can be performed in a mathematical programming language like Fortran or using a statistical software package such as SPSS or SAS. At activity 6322, variables identified in any 3 of the 4 techniques of activity 6321 as “non-contributing” and/or “insignificant” can be removed. Although in some situations a quantity of 100 or less variables can be ideal in terms of computational efficiency, an arbitrary cut-off generally should not be forced to ensure the number of variables used in the dataset is 100 or less.
Referring to
At activity 6500, a determination can be made regarding whether cluster weights are needed to meet marketing coverage or marketing efficiency objectives. At activity 6510, if no weights are needed, Method 6 can proceed to activity 6600. At activity 6520, if weights are needed, each cluster can be assigned a weight using the decision rule of one of activities 6520, 6522, and 6523. In any event, if weights are needed, each cluster must be assigned a weight regardless of the cluster's magnitude. At activity 6521, if the cluster is to be unweighted, the weight can be set as w=1. At activity 6522, if the cluster is to be overweighted, the weight can be set as w>1. At activity 6523, if the cluster is to be underweighted, the weight can be set as 0<w<1.
At activity 6600, activities 5300 through 5450 of Method 5 can be executed. At activity 6700, Method 6 can be terminated.
At activity 7100, a use for panel data is ascertained. For example, panel data can be used as an objective measure of behavior that can be input into a clustering technique such as that of Method 2. As another example, panel data can be used for post-clustering analyses, e.g. tracking, promotion, media performance, or positioning.
At activity 7110, if the panel data is to be used as objective measure, then at activity 7111, the data collection instrument, e.g. survey, can be refined using Method 1 (i.e., the Babbitt Score technique). At activity 7112, the data collection instrument can be fielded within a panel (e.g. IRI, Nielsen, Simmons, comSCORE and/or MRI). At activity 7113, observations can be extracted from the collected data to assemble a dataset that reflects a category's and/or industry's underlying demographics. In some situations, this activity can be necessary to ensure that the clustering-solution developed using this dataset is truly representative of a market or industry and is not just the function of an idiosyncratic group of the overall population.
At activity 7114, the dataset can be obtained for the extracted observations, the dataset having variables, possible values, and provided values. At activity 7115, the panel-based behavioral variables can be appended to each corresponding observation in the dataset. At activity 7116, any panel variables that are continuous can be transformed into categorical or scalar variables. This transformation can be performed by analyzing distribution boundaries within a series of ranges to find the boundaries that create as normal distribution as possible. In many situations, linear optimization is the most efficient method for executing this analysis.
At activity 7117, the dataset from activity 7116 can be input into activity 2420 of Method 2 (the Bestfit Clustering technique) and clustering can proceed using Method 2. At activity 7118 the process can terminate.
Turning now to use of the panel data for post-lustering analysis, at activity 7121, data is collected. At activity 7121.1, if the dataset was developed using the panel data as an objective measure of behavior, Method 7 can continue to activity 7122, where Methods 2 through 4 can be executed.
Alternatively, if panel data was used for post-clustering analysis, Method 7 can continue to activity 7121.2, where the data collection instrument can be refined using Method 1 (the Babbitt Score technique). At activity 7123, Methods 2 through 5 can be executed. At activity 7123.1, the typing tool developed in activity 7123 can be used to cluster score a representative sample of the panel's members using an expedient contact channel (e.g., outbound telephone, e-mail/electronic surveys, and/or mail-based surveys, etc.). At activity 7124, a cluster level analysis can be executed using the panel data. At activity 7125, Method 7 can be terminated.
At activity 8100, a pilot survey can be developed and fielded. At activity 8200, the survey can be refined. This refinement can implement Method 1 and/or Method 2.
At activity 8300, a full survey can be fielded. At activity 8400, the data from the survey can be cleaned, refined, and otherwise analyzed, using, for example, Method 4. At activity 8500, clusters can be created using, for example, Method 1, 3, and/or 7. At activity 8600, clusters can be refined using, for example, Method 4.
At activity 8700, Method 5 can be implemented. At activity 8750, Method 6 can be used. At activity 8800, panel variables can be appended to the results of activity 8750, and Method 7 can then be utilized in activity 8850. At activity 8900, insights into the segments can be gained, and at activity 8950, marketing ideas, strategies, and tactics can be developed and implemented.
Information device 9 can include well-known components such as one or more processors 9120, one or more memories 9140 containing instructions 9160, one or more input/output (I/O) devices 9180, and one or more network interfaces 9190.
In one embodiment of information device 9, each processor 9120 can be a general purpose microprocessor, such a the Pentium series microprocessor manufactured by the Intel Corporation of Santa Clara, Calif. In another embodiment, the processor can be an Application Specific Integrated Circuit (ASIC) which has been designed to implement in its hardware and/or firmware at least a part of a method in accordance with an embodiment of the present invention.
Any memory 9140 can be coupled to a processor 9120 and can store instructions 9160 adapted to be executed by processor 9120 according to one or more actions of methods 1 through 9. Memory 9140 can be any device capable of storing analog or digital information, such as a hard disk, Random Access Memory (RAM), Read Only Memory (ROM), flash memory, a compact disk, a magnetic tape, a floppy disk, and any combination thereof.
Instructions 9160 can be embodied in software, which can take any of numerous forms that are well-known in the art. For example, information device 9 can access one or more databases having a flat file or a relational organization, and a centralized or distributed architecture. For instance, those of skill in the art can tailor items such as an SQL database to provide the functionality of methods 1 through 8 and 10. One supplier of such database items can be Oracle Corporation, of Redwood Shores, Calif. Moreover, software tools such as EDI, FTP, HTTP, HTML, XML, cXML, XSL, and WAP can be utilized for communications between information devices. Additionally, information device 9 can utilize platform-independent and/or network-centric software tools such as, for example, Java or JavaScript.
Any input/output (I/O) device 9180 can be an audio and/or visual device, including, for example, a monitor, display, keyboard, keypad, touchpad, pointing device, microphone, speaker, video camera, camera, scanner, and/or printer, including a port to which an I/O device can be attached or connected.
Any network interface 9190 can be a telephone, a traditional data modem, a fax modem, a cable modem, a digital subscriber line interface, a bridge, a hub, a router, or other similar devices.
At activity 10020, the modes, their corresponding variable identifications, and their corresponding cluster identifications are stored. At activity 10030, modes of given values for all variables in {K} for cluster n+y are calculated. At activity 10040, the modes, their corresponding variable identifications, and their corresponding cluster identifications are stored. At activity 10050, an assessment is made regarding the number of clusters for which modes have been calculated. At activity 10060, if the number of clusters for which modes have been calculated is equal to N, Method 10 proceeds to activity 10080. At activity 10070, if the number of clusters for which modes have been calculated is less than N, Method 10 returns to activity 10030.
At activity 10080, for each cluster, the value provided by each constituent member for variable k is compared to the cluster's mode for variable k. At activity 10090, the value of i_{n }for k is compared to the mode of k_{n}, where i is an element of {R}, which is an element of {I}, and R consists of the set of observations, whose total number of constituents is greater than 0, and I is the domain of integers, and i_{n }is a member of cluster n, and k_{n }is the variable k as answered by cluster n.
At activity 10100, if the value of in for k is equal to the mode of k_{n}, V_{kni }is set to 1, where V_{kni }is the score for an observation i that is in cluster n and has provided answer value V for variable k. At activity 10110, if the value of i_{n }for k is not equal to the mode of k_{n}, V_{kni }is set to zero.
At activity 10120, V_{kni }is stored. At activity 10130, V_{kni }can be adjusted by an indicated weight. At activity 10140, if a weight was specified, V_{kni }is multiplied by the corresponding weight for k. At activity 10150, if a weight was not specified, V_{kni }is multiplied by 1. At activity 10160, V_{kni }is stored. At activity 10170, activities 10080 to 10160 are repeated until a score of V is calculated for all observations in their respective clusters for all variables k. At activity 10200, all scores V are summed for all observations across all variables to arrive at the fitness score. At activity 10300, the fitness is stored. At activity 10400, method 10 is terminated.
Additional Embodiments
Still other advantages of the present invention will become readily apparent to those skilled in this art from the above-recited detailed description. Accordingly, the drawings and descriptions are to be regarded as illustrative in nature, and not as restrictive. For example, embodiments of Methods 1, 2, 3, 4, 5 6, 7, 8 and/or 10 of the present invention can be viewed as germane to a number of disparate disciplines beyond business and marketing. The following list outlines examples of problems in these fields that can be solved by applying one or more of the aforementioned methods.
Cited Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|
US5832182 * | Apr 24, 1996 | Nov 3, 1998 | Wisconsin Alumni Research Foundation | Method and system for data clustering for very large databases |
US6003029 * | Aug 22, 1997 | Dec 14, 1999 | International Business Machines Corporation | Automatic subspace clustering of high dimensional data for data mining applications |
US6374251 * | Mar 17, 1998 | Apr 16, 2002 | Microsoft Corporation | Scalable system for clustering of large databases |
US6615205 * | Dec 22, 2000 | Sep 2, 2003 | Paul M. Cereghini | Horizontal implementation of expectation-maximization algorithm in SQL for performing clustering in very large databases |
US6745184 * | May 31, 2001 | Jun 1, 2004 | Rosetta Marketing Strategies Group | Method and system for clustering optimization and applications |
US20020013752 * | Dec 14, 2000 | Jan 31, 2002 | Johnson Christopher D. | Rapid valuation of portfolios of assets such as financial instruments |
US20030110112 * | Dec 14, 2000 | Jun 12, 2003 | Johnson Christopher D. | Methods and systems for automated inferred valuation of credit scoring |
Reference | ||
---|---|---|
1 | * | Johnson et al. "Effects of Question Context and Response Order on Attitude Questions", Proceedings of The Survey Research Methods Section, American Statistical Msociation, 1998, pp. 857-860. |
2 | * | Trindade, David, Survey Data: Use of Scatter Plots for Displaying Scale and Consistency Factorse, Proceeding of The Survey Research Methods Section, American Statistical Association, 1995, pp. 99-101. |
Citing Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|
US7243056 | Feb 21, 2002 | Jul 10, 2007 | Honda Research Institute Europe Gmbh | Strategy parameter adaptation in evolution strategies |
US7333960 | Jul 30, 2004 | Feb 19, 2008 | Icosystem Corporation | Methods and systems for applying genetic operators to determine system conditions |
US7356518 * | Aug 20, 2004 | Apr 8, 2008 | Icosystem Corporation | Methods and systems for multi-participant interactive evolutionary computing |
US7363281 * | Jan 24, 2005 | Apr 22, 2008 | Honda Research Institute Europe Gmbh | Reduction of fitness evaluations using clustering techniques and neural network ensembles |
US7428514 | Jan 11, 2005 | Sep 23, 2008 | Honda Research Institute Europe Gmbh | System and method for estimation of a distribution algorithm |
US7444309 | Oct 28, 2002 | Oct 28, 2008 | Icosystem Corporation | Method and system for implementing evolutionary algorithms |
US7603326 | May 8, 2006 | Oct 13, 2009 | Icosystem Corporation | Methods and systems for interactive evolutionary computing (IEC) |
US7624077 | Aug 28, 2007 | Nov 24, 2009 | Icosystem Corporation | Methods and systems for multi-participant interactive evolutionary computing |
US7707220 | Sep 29, 2006 | Apr 27, 2010 | Icosystem Corporation | Methods and apparatus for interactive searching techniques |
US7783583 | Sep 6, 2006 | Aug 24, 2010 | Honda Research Institute Europe Gmbh | Evolutionary search for robust solutions |
US7792816 | Jan 31, 2008 | Sep 7, 2010 | Icosystem Corporation | Method and system for fast, generic, online and offline, multi-source text analysis and visualization |
US7856434 | Nov 12, 2007 | Dec 21, 2010 | Endeca Technologies, Inc. | System and method for filtering rules for manipulating search results in a hierarchical search and navigation system |
US7882048 | Jan 15, 2008 | Feb 1, 2011 | Icosystem Corporation | Methods and systems for applying genetic operators to determine system conditions |
US7908159 * | Feb 12, 2003 | Mar 15, 2011 | Teradata Us, Inc. | Method, data structure, and systems for customer segmentation models |
US7912823 | Oct 31, 2007 | Mar 22, 2011 | Endeca Technologies, Inc. | Hierarchical data-driven navigation system and method for information retrieval |
US8019752 | Nov 10, 2005 | Sep 13, 2011 | Endeca Technologies, Inc. | System and method for information retrieval from object collections with complex interrelationships |
US8117139 | Oct 13, 2009 | Feb 14, 2012 | Icosystem Corporation | Methods and systems for interactive evolutionary computing (IEC) |
US8117140 | Jan 26, 2011 | Feb 14, 2012 | Icosystem Corporation | Methods and systems for applying genetic operators to determine systems conditions |
US8171022 * | Nov 7, 2005 | May 1, 2012 | Johnston Jeffrey M | Methods, systems, and computer program products for facilitating user interaction with customer relationship management, auction, and search engine software using conjoint analysis |
US8423323 | Sep 21, 2006 | Apr 16, 2013 | Icosystem Corporation | System and method for aiding product design and quantifying acceptance |
US8489532 | Mar 13, 2012 | Jul 16, 2013 | Information Resources, Inc. | Similarity matching of a competitor's products |
US8504598 | Jan 28, 2008 | Aug 6, 2013 | Information Resources, Inc. | Data perturbation of non-unique values |
US8548937 | Aug 17, 2011 | Oct 1, 2013 | Wisercare Llc | Medical care treatment decision support system |
US8676802 | Nov 30, 2006 | Mar 18, 2014 | Oracle Otc Subsidiary Llc | Method and system for information retrieval with clustering |
US8719266 | Jul 22, 2013 | May 6, 2014 | Information Resources, Inc. | Data perturbation of non-unique values |
US8818838 | Aug 26, 2009 | Aug 26, 2014 | Henry Rak Consulting Partners, LLC | System and method for efficiently developing a hypothesis regarding the structure of a market |
US8992446 * | Jun 20, 2010 | Mar 31, 2015 | Holland Bloorview Kids Rehabilitation Hospital | Procedure for denoising dual-axis swallowing accelerometry signals |
US9026591 | May 25, 2011 | May 5, 2015 | Avaya Inc. | System and method for advanced communication thread analysis |
US9058560 | Feb 13, 2012 | Jun 16, 2015 | Superior Edge, Inc. | Methods, apparatus and systems for generating, updating and executing an invasive species control plan |
US9113590 | Aug 6, 2012 | Aug 25, 2015 | Superior Edge, Inc. | Methods, apparatus, and systems for determining in-season crop status in an agricultural crop and alerting users |
US20020051020 * | Sep 21, 2001 | May 2, 2002 | Adam Ferrari | Scalable hierarchical data-driven navigation system and method for information retrieval |
US20020099929 * | Nov 9, 2001 | Jul 25, 2002 | Yaochu Jin | Multi-objective optimization |
US20020165703 * | Feb 21, 2002 | Nov 7, 2002 | Markus Olhofer | Strategy parameter adaptation in evolution strategies |
US20040117366 * | Dec 12, 2002 | Jun 17, 2004 | Ferrari Adam J. | Method and system for interpreting multiple-term queries |
US20040210545 * | Oct 28, 2002 | Oct 21, 2004 | Juergen Branke | Method and system for implementing evolutionary algorithms |
US20050038781 * | Sep 8, 2003 | Feb 17, 2005 | Endeca Technologies, Inc. | Method and system for interpreting multiple-term queries |
US20050118612 * | Jul 30, 2004 | Jun 2, 2005 | Icosystem Corporation | Methods and systems for applying genetic operators to determine system conditions |
US20050119983 * | Aug 20, 2004 | Jun 2, 2005 | Eric Bonabeau | Methods and systems for multi-participant interactive evolutionary computing |
US20050209982 * | Jan 24, 2005 | Sep 22, 2005 | Yaochu Jin | Reduction of fitness evaluations using clustering techniques and neural network ensembles |
US20050256684 * | Jan 11, 2005 | Nov 17, 2005 | Yaochu Jin | System and method for estimation of a distribution algorithm |
US20060041401 * | Aug 12, 2005 | Feb 23, 2006 | Johnston Jeffrey M | Methods, systems, and computer program products for facilitating user choices among complex alternatives using conjoint analysis in combination with psychological tests, skills tests, and configuration software |
US20060053104 * | Nov 8, 2005 | Mar 9, 2006 | Endeca Technologies, Inc. | Hierarchical data-driven navigation system and method for information retrieval |
US20060195204 * | May 8, 2006 | Aug 31, 2006 | Icosystem Corporation | Methods and Systems for Interactive Evolutionary Computing (IEC) |
US20070025540 * | Jul 7, 2005 | Feb 1, 2007 | Roger Travis | Call center routing based on talkativeness |
US20070036323 * | Jan 26, 2006 | Feb 15, 2007 | Roger Travis | Call center routing |
US20070067279 * | Sep 29, 2006 | Mar 22, 2007 | Icosystem Corporation | Methods and Apparatus for Interactive Searching Techniques |
US20070094167 * | Sep 6, 2006 | Apr 26, 2007 | Bernhard Sendhoff | Evolutionary search for robust solutions |
US20070106658 * | Nov 10, 2005 | May 10, 2007 | Endeca Technologies, Inc. | System and method for information retrieval from object collections with complex interrelationships |
US20070233730 * | Nov 7, 2005 | Oct 4, 2007 | Johnston Jeffrey M | Methods, systems, and computer program products for facilitating user interaction with customer relationship management, auction, and search engine software using conjoint analysis |
US20070282812 * | Mar 8, 2007 | Dec 6, 2007 | Superior Edge, Inc. | Process execution support system |
US20080016093 * | Jul 11, 2006 | Jan 17, 2008 | Clement Lambert Dickey | Apparatus, system, and method for subtraction of taxonomic elements |
US20080021855 * | Aug 28, 2007 | Jan 24, 2008 | Icosystem Corporation | Methods And Systems For Multi-Participant Interactive Evolutionary Computing |
US20080021897 * | Jul 19, 2006 | Jan 24, 2008 | International Business Machines Corporation | Techniques for detection of multi-dimensional clusters in arbitrary subspaces of high-dimensional data |
US20080140374 * | Jan 15, 2008 | Jun 12, 2008 | Icosystem Corporation | Methods and Systems for Applying Genetic Operators to Determine System Conditions |
US20080263000 * | Jan 28, 2008 | Oct 23, 2008 | John Randall West | Utilizing aggregated data |
US20080263065 * | Jan 28, 2008 | Oct 23, 2008 | John Randall West | Data perturbation of non-unique values |
US20080319829 * | Jan 28, 2008 | Dec 25, 2008 | Herbert Dennis Hunt | Bias reduction using data fusion of household panel data and transaction data |
US20090018996 * | Jan 28, 2008 | Jan 15, 2009 | Herbert Dennis Hunt | Cross-category view of a dataset using an analytic platform |
US20090144617 * | Jan 31, 2008 | Jun 4, 2009 | Pablo Funes | Method and system for fast, generic, online and offline, multi-source text analysis and visualization |
US20090248427 * | Mar 25, 2008 | Oct 1, 2009 | Jerome Dale Johnson | Initiative designer |
US20090248466 * | Sep 8, 2008 | Oct 1, 2009 | Jerome Dale Johnson | Initiative analyzer |
US20120271872 * | Oct 25, 2012 | Tom Chau | Procedure for denoising dual-axis swallowing accelerometry signals | |
US20130110584 * | Oct 26, 2012 | May 2, 2013 | Global Market Insite, Inc. | Identifying people likely to respond accurately to survey questions |
U.S. Classification | 1/1, 706/48, 706/52, 705/19, 706/45, 707/999.006, 707/999.101 |
International Classification | G06Q30/00, G06F17/30 |
Cooperative Classification | Y10S707/99942, Y10S707/99936, G06Q30/02, G06Q20/207 |
European Classification | G06Q30/02, G06Q20/207 |
Date | Code | Event | Description |
---|---|---|---|
May 31, 2001 | AS | Assignment | Owner name: ROSETTA MARKETING STRATEGIES GROUP, NEW JERSEY Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:CHOI, LAWRENCE J.;KUENNE, CHRISTOPHER B.;HOLSTEIN, II., KURT E.;REEL/FRAME:011883/0143 Effective date: 20010529 |
Jul 11, 2005 | AS | Assignment | Owner name: RMSG LLC, NEW JERSEY Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:ROSETTA MARKETING STRATEGIES GROUP;REEL/FRAME:016509/0515 Effective date: 20050707 |
Nov 8, 2005 | AS | Assignment | Owner name: BROWN BROTHERS HARRIMAN & CO., NEW YORK Free format text: SECURITY INTEREST;ASSIGNOR:RMSG, LLC, A DELAWARE LIMITED LIABILITY COMPANY;REEL/FRAME:017186/0556 Effective date: 20050825 |
Jul 24, 2008 | AS | Assignment | Owner name: BROWN BROTHERS HARRIMAN & CO., AS AGENT, NEW YORK Free format text: SECURITY AGREEMENT;ASSIGNOR:BROWN BROTHERS HARRIMAN & CO.;REEL/FRAME:021291/0001 Effective date: 20080723 |
Feb 9, 2009 | FPAY | Fee payment | Year of fee payment: 4 |
Sep 20, 2010 | AS | Assignment | Owner name: BANK OF AMERICA, N.A., AS ADMINISTRATIVE AGENT, IL Free format text: SECURITY AGREEMENT;ASSIGNOR:RMSG LLC;REEL/FRAME:025008/0225 Effective date: 20100920 |
Sep 21, 2010 | AS | Assignment | Owner name: RMSG LLC, NEW JERSEY Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:BROWN BROTHERS HARRIMAN & CO;REEL/FRAME:025017/0498 Effective date: 20100920 |
Jul 5, 2011 | AS | Assignment | Owner name: RMSG LLC, NEW JERSEY Free format text: RELEASE OF SECURITY INTEREST;ASSIGNOR:BANK OF AMERICA, N.A., AS ADMINISTRATIVE AGENT;REEL/FRAME:026544/0885 Effective date: 20110701 |
Mar 25, 2013 | REMI | Maintenance fee reminder mailed | |
Aug 9, 2013 | LAPS | Lapse for failure to pay maintenance fees | |
Aug 9, 2013 | REIN | Reinstatement after maintenance fee payment confirmed | |
Oct 1, 2013 | FP | Expired due to failure to pay maintenance fee | Effective date: 20130809 |
Sep 8, 2014 | PRDP | Patent reinstated due to the acceptance of a late maintenance fee | Effective date: 20140910 |
Sep 10, 2014 | SULP | Surcharge for late payment | |
Sep 10, 2014 | FPAY | Fee payment | Year of fee payment: 8 |
Oct 7, 2015 | AS | Assignment | Owner name: ROSETTA MARKETING GROUP, LLC, NEW JERSEY Free format text: MERGER;ASSIGNOR:RMSG LLC;REEL/FRAME:036745/0569 Effective date: 20111220 |